The polls: public opinion and Secretary of Defense Donald Rumsfeld.

Panagopoulos, Costas

George W. Bush's secretary of defense, Donald Rumsfeld, has been characterized as a "central political figure in our time." (1) He is also one of the most controversial and contentious cabinet members of the Bush administration. Analysts argue Rumsfeld is simultaneously "widely admired" and "ever so widely resented," even as critics rile charges of failed leadership against him. (2) Republican Senator John McCain of Arizona, a senior member of the Senate Armed Services Committee, which has oversight of military operations and considerable influence over the Pentagon budget, has publicly stated he has "no confidence in the defense secretary." (3) Despite these criticisms, Rumsfeld, who has been described as "one of the most high-profile, powerful and polarizing Defense secretaries since Robert McNamara in the 1960s," remains a leading advisor to the president and was tapped by President Bush for an extended tour of duty as secretary of defense in his second administration. (4)

The events following the terrorist attacks on September 11, 2001, including the wars in Afghanistan and Iraq that ensued, have catapulted Secretary Rumsfeld into the national spotlight. This attention has enabled Americans to monitor and assess his performance as secretary of defense. This article investigates the public's evaluations of Secretary Rumsfeld's job performance over time. It also seeks to explain the patterns in opinion we observe. More generally, this study assesses the degree to which the public's evaluations of key cabinet members, such as Donald Rumsfeld, impact public evaluations of the president.

As a theoretical matter, we may expect that presidents, as rational actors, may act to maximize their own popularity by replacing unpopular cabinet members, but this is reasonable only if there exists a relationship between cabinet member performance ratings and presidential approval. Lacking evidence of such a direct link, it is not clear that presidents stand to benefit much from dismissing cabinet secretaries, even if they are unpopular. An alternative possibility is that presidents will not mind unpopular subordinates as their lack of popularity may deflect criticism away from the president and onto the subordinate, distancing the president from direct fallout. In both of these scenarios, however, unpopular subordinates are not, at a minimum, adversely affecting public evaluations of presidential performance. Under such circumstances, subordinates are expected to survive, whereas a president may have incentives to dismiss unpopular administration officials whose poor approval ratings hurt the president's directly. This may help to explain why Secretary Rumsfeld, for example, remains a central figure in the Bush administration despite calls for his resignation. A growing literature reflects on conditions under which principals (prime ministers, presidents) may sack agents (cabinet members) (Palmer 1995), and this study aims to present an examination of the U.S. case.

Public Opinion and Donald Rumsfeld

The American public's attitudes about Donald Rumsfeld have been regularly assessed by polling organizations during his tenure as secretary of defense in the George W. Bush administrations. This article examines the dynamics of public opinion toward Rumsfeld between October 2001 and January 2005. (5) Quarterly data on Rumsfeld's job approval ratings are displayed in Figure 1. The data presented reveal variation in job approval between 2001 and 2005 and also show that the public's evaluation of Rumsfeld's performance as secretary of defense has declined steadily during this period. Rumsfeld's job approval rating, which exceeded 90 percent in the immediate aftermath of the September 11, 2001 attacks, has hovered around 50 percent in the four most recent quarters included in the analysis. In fact, Rumsfeld's job approval has declined in every quarter except two over this time period. For the first time in the series, a majority of Americans indicated they disapproved of Rumsfeld's job performance during the fourth quarter of 2004. Rumsfeld's approval ratings have ranged from a minimum of 47 percent to a maximum of 93 percent over this time period, and Rumsfeld's average quarterly job approval rating has been 68.9 percent. Regressing Rumsfeld's approval ratings on time and a constant, the data indicate Rumsfeld's job approval has declined by almost twelve percentage points every year between 2001 and 2005 (ordinary least squares regression coefficient = -3.6; standard error = 0.20; p < 0.01; R-squared = 0.96; N = 14). The empirical evidence confirms the American public's confidence in Secretary Rumsfeld's ability to handle his duties has deteriorated significantly in recent years.

[FIGURE 1 OMITTED]

Presidential and Rumsfeld Approval

Given the public's weakening assessments about Secretary Rumsfeld's job performance, many analysts suggested he was a liability to President Bush both during the 2004 presidential election and in the second Bush administration that followed it. Intense scrutiny and criticism directed toward Secretary Rumsfeld during the prisoner abuse scandal at Abu Ghraib in particular fueled calls for his resignation or dismissal during the spring of 2004. (6) To date, however, no study of which I am aware has examined empirically the relationship between public opinion about Secretary Rumsfeld and President Bush. To what extent does public sentiment about Secretary Rumsfeld's job performance impact assessments of presidential job performance? Moreover, is President Bush's job approval related to Secretary Rumsfeld's approval? To determine whether public evaluations of the president and his defense secretary are interdependent or independent, I have assembled quarterly approval measures for President Bush for the same period spanning 2001 to 2005. (7) Presidential approval was determined using data collected by the Gallup organization.

In addition to quarterly job approval ratings for Secretary Rumsfeld, Figure 1 also displays job approval ratings for President Bush for the corresponding quarterly periods between 2001 and 2005. The data show that the public's approval of Secretary Rumsfeld's job performance has generally exceeded the president's job approval for most of this period. President Bush's approval ratings surpassed Secretary Rumsfeld's approval only starting during the third quarter of 2004, even as President Bush's job approval has remained higher than Secretary Rumsfeld's since.

Figure 1 also suggests a close association between the approval measures for the two political actors over this period. An analysis of the data reveals a high instantaneous (bivariate) correlation between the two series (Pearson's R correlation coefficient = 0.96; p < 0.01). This initial evidence suggests a strong interdependent relationship between Secretary Rumsfeld's job approval and President Bush's performance evaluations.

Correlation does not necessarily signify causation, however, and the relationship we observe may not be causal. Moreover, causality, if it exists, can flow in either direction. In other words, President Bush's job approval rating may influence Secretary Rumsfeld's ratings or vice versa. To investigate these possibilities and to determine causality, I use Granger (1969) causality tests. Granger causality tests whether lagged information on a variable Y provides any statistically significant information about a variable X in the presence of lagged X. If not, then "Y does not Granger-cause X." I estimate regressions with one lag to determine whether Rumsfeld approval Granger-causes Bush approval and vice versa. Given that approval is measured quarterly, I did not extend the analysis beyond one lag.

Table 1 presents the results of the causality tests. The findings indicate no causal relationship between Rumsfeld approval and Bush approval. In both equations, only the lagged dependent variable influences the dependent variable and is statistically significant at conventional levels. The inclusion of lagged Rumsfeld approval does not help to predict Bush approval, for instance, and lagged Bush approval is similarly unrelated to Rumsfeld approval ratings. More importantly, the results of F-tests on the other series' coefficients displayed in Table 1 never reach the critical values needed to reject the null hypothesis that one variable does not cause the other. In other words, Secretary Rumsfeld's job approval appears to be independently derived from President Bush's approval, and President Bush's approval is not affected directly by Secretary Rumsfeld's approval ratings. The results of this analysis suggest each of these two political actors derive their job approval measures independently.

Explaining Rumsfeld Approval

The task remains to explain the variation we observe in Rumsfeld's approval ratings over this period. What forces influence the public's assessment of Secretary Rumsfeld's job performance? Following Burden and Mughan (1999), I establish three general categories of explanatory variables: general political attitudes, economic variables, and patterns of media coverage. Details on operationalization appear in the Appendix.

Scholars have demonstrated that media coverage influences presidential approval (Burden and Mughan 1999; Ragsdale 1997). I expect that the impact of media coverage, both quantity and tone, extends to the secretary of defense. My measure of the amount of media coverage is the quarterly count of stories about Secretary Rumsfeld that have appeared in four national newspapers over the period of this study. (8) To gauge the tone of the coverage, I group each of these stories into two categories: non-scandal and scandal. I presume that all scandal stories will be negative in tone with respect to Secretary Rumsfeld. I also measure the total number of stories that have appeared in the same outlets that discussed the number of American casualties or deaths in either of the two wars undertaken during this period (Iraq and Afghanistan). I expect that greater coverage of war casualties will negatively impact Secretary Rumsfeld's performance evaluations.

Figures 2 and 3 display the quarterly counts of each of these categories of newspaper stories. The data presented in Figure 2 show that the number of non-scandal stories about Rumsfeld that appeared in these outlets ranged from a low of 484 to a high of 1,403. The total number of stories about war casualties ranged quarterly from 290 to 1,309. An analysis of these data reveals no discernible trends in the amount of coverage in each of these categories during the period of this study. Figure 3 displays the total number of scandal stories per quarter about Secretary Rumsfeld that appeared in the newspaper outlets over the same period. One clearly notices the spike in the total number of scandal stories about Rumsfeld that follows the prisoner abuse scandal at Abu Ghraib during the spring of 2004. Up until that time, the national newspaper outlets included in the study rarely published articles that connected Secretary Rumsfeld to scandals. A total of 118 such stories appeared in these outlets in the second quarter of 2004, however. Overall, the total number of scandal stories about Rumsfeld over the entire period increased significantly at an average rate of about three stories per quarter. This finding may present some initial insight to explain the decline in Rumsfeld approval we observe.

[FIGURES 2-3 OMITTED]

The second set of explanatory variables is economic. I consider two measures of macroeconomic performance: quarterly change in unemployment and inflation. While economic variables have been shown to influence presidential evaluations, there is scant evidence of a similar direct effect for cabinet members or, in this case, for Secretary Rumsfeld's approval ratings. It is conceivable, however, that economic indicators, viewed through the prism of war, may be conflated with evaluations about military performance and spill over to affect Rumsfeld's approval ratings.

The main attitudinal variable in the analysis is macropartisanship. MacKuen, Erikson, and Stimson (1989) have demonstrated that partisan identification can be dynamic and can affect--and be affected by--changes in the political surroundings. Following Burden and Mughan (1999, 243) it is measured as the proportion of partisan (Republican) identifiers, and it is included to take account of long-standing partisan predispositions in the electorate that may influence Secretary Rumsfeld's approval.

Table 2 presents descriptive statistics on the range of variables included in the analysis that follows.

To estimate the linear model while correcting for serial correlation, I use the Prais-Winsten regression technique. The results of the estimation are presented in Table 3. The findings indicate that Secretary Rumsfeld's job performance evaluations are not influenced by the economic or attitudinal variables, sets of indicators routinely shown to affect presidential approval. Only media coverage influences the public's assessment of Secretary Rumsfeld's job performance. Non-scandal stories about Rumsfeld positively impact his overall approval. The coefficient indicates that for each additional non-scandal story, Rumsfeld's job approval will increase by 0.07 percentage points. The other two media variables also exert the anticipated impact. More media stories that associate Rumsfeld with scandal depress his overall approval, as do more stories about war casualties. For each additional scandal story that appears in the newspaper, Rumsfeld's approval declines by 0.17 percentage points. Similarly, for each additional story about war casualties, the secretary's performance rating will decline by 0.05 percentage points. These findings confirm expectations about the effects of each media coverage variable.

Conclusions

The results of this analysis suggest that Secretary Rumsfeld's job approval ratings are independent from President Bush's performance evaluations and vice versa. Rumsfeld's job approval is driven by media coverage and not by the forces of macropartisanship and national economic performance.

On a more general note, the results of these analyses suggest that Americans are quite sophisticated with respect to ascribing accountability for government performance to specific political actors. Americans appear to hold the secretary of defense responsible for matters within his purview or domain, but not necessarily beyond that. The condition of the economy, for example, does not appear to affect Americans' assessments of the performance of the secretary of defense. Additional research may reveal similar patterns for other members of a president's cabinet, lending additional support for the notion that Americans attribute responsibility for specific failures or accomplishments within specific policy domains to the appropriate executive branch official.

Generalizations from this study may be hazardous given Secretary Rumsfeld's unusual public presence and visibility. Few cabinet members attract the attention or generate the volume of media coverage Rumsfeld has either enjoyed or endured. Nevertheless, we may speculate that presidential job approval may be insulated from the deleterious effect of unpopular cabinet officials. This may help us to understand why presidents retain cabinet members, even in the face of deteriorating approval ratings.

Appendix

Rumsfeld approval. Job approval ratings for Secretary of Defense Donald Rumsfeld were collected from thirty-seven nationally representative opinion surveys conducted between October 1, 2001 and March 31, 2005. Survey organizations included: Gallup, Gallup/CNN/USA Today, ABC/Washington Post, Fox News/Opinion Dynamics, Quinnipiac University Poll, Time/SRBI, CBS News, and Time/CNN/Harris. Details are available upon request. In most surveys, the wording for the job approval question was consistently: "Do you approve or disapprove of the way Donald Rumsfeld is handling his job as Secretary of Defense?" In surveys conducted by Fox News, question wording was: "Do you approve or disapprove of the job Donald Rumsfeld is doing as Secretary of Defense?" In surveys conducted by Time/CNN/Harris, question wording was: "In general, do you approve or disapprove of the way Donald Rumsfeld is handling his job as Secretary of Defense?" Responses other than "approve" and "disapprove" were excluded before computing overall percentages. The measure of "relative approval" is akin to Stimson's (1976) indicator.

Bush approval. Job approval data for President Bush are computed in the same manner as approval for Secretary Rumsfeld (aggregated quarterly) as indicators of "relative approval" in order to be comparable. Data are compiled from surveys conducted by the Gallup organization. Question wording is: "Do you approve or disapprove of the way President Bush is handling his job as President?"

Economic variables: unemployment/inflation. Data for these variables were obtained from the Bureau of Labor Statistics. Measures represent quarterly change in overall levels of unemployment and inflation during the period of the study.

Macropartisanship. Proportion of all party identifiers in CBS News/New York Times polls (dates) or Gallup surveys (dates) who are Republican: % Republican/(% Republican + % Democrat). Independents were excluded.

Media coverage (newspaper).

Rumsfeld/non-scandal: The total number of stories (aggregated quarterly) that mention Secretary Donald Rumsfeld that appeared in the New York Times, Chicago Tribune, Los Angeles Times, or Washington Post.

Rumsfeld/scandal: The total number of stories (aggregated quarterly) that mention Secretary Donald Rumsfeld in association with a scandal that appeared in the New York Times, Chicago Tribune, Los Angeles Times, or Washington Post.

War casualties stories: The total number of stories (aggregated quarterly) that mention war casualties and/or the deaths of American soldiers in conflicts in either Afghanistan or Iraq that appeared in the New York Times, Chicago Tribune, Los Angeles Times, or Washington Post. These data were acquired through Lexis/Nexis.

References

Burden, Barry C., and Anthony Mughan. 1999. Public opinion and Hillary Clinton. Public Opinion Quarterly 63: 237-50.

Granger, C. W. J. 1969. Investigating causal relations by econometric methods and cross-spectral methods. Econometrica 34: 424-38.

MacKuen, Michael, Robert S. Erikson, and James A. Stimson. 1989. Macropartisanship. American Political Science Review 83:1125-42.

Palmer, Matthew. 1995. Toward an economics of comparative political organization: Examining ministerial responsibility. Journal of Law, Economics, & Organization 11 : 164-88.

Ragsdale, Lyn. 1997. Disconnected politics: Public opinion and presidents. In Understanding Public Opinion, edited by Barbara Norrander and Clyde Wilcox. Washington, DC: Congressional Quarterly Press.

Stimson, James A. 1976. Public support for American presidents: A cyclical model. Public Opinion Quarterly 40: 1-21.

(1.) William F. Buckley, "Understanding Rumsfeld," National Review, October 24, 2003.

(2.) Ibid.

(3.) Associated Press, December 14, 2004.

(4.) Richard W. Stevenson and Thom Shanker, "Bush Sticks with Rumsfeld," New York Times, December 4, 2004.

(5.) No single time series is available for this analysis. I have pooled thirty-two nationally representative surveys that asked respondents whether or not they approve of Donald Rumsfeld's handling of his job as secretary of defense. Full descriptions of the samples, dates, and survey organizations appear in the Appendix. Data presented in Figure 1 represent quarterly aggregations of "relative approval" (see Stimson [1976] and Burden and Mughan [ 1999] for a discussion of this operationalization. Approval is measured as: percent approve/[percent approve + percent disapprove]). In three cases, job approval data were unavailable (first quarter 2004, fourth quarter 2002, and second quarter, 2002). In cases where Rumsfeld approval data were missing for a particular quarter (at time t), the value was imputed with the mean of the observations at t - 1 and t + 1 (see Burden and Mughan 1999). When the approval question is asked multiple times within a single quarter, the observations are averaged to create quarterly observations.

(6.) Elisabeth Bumiller and Richard Stevenson, "Rumsfeld Chastised by President for His Handling of Iraq Scandal," New York Times, May 6, 2004.

(7.) Presidential job approval data are determined exactly as Rumsfeld's job approval ratings are, thereby making these two measures comparable over the complete time period. There were no missing data for presidential approval.

(8.) The newspapers are the Chicago Tribune, New York Times, Los Angeles Times, and Washington Post.

COSTAS PANAGOPOULOS

Yale University

Costas Panagopoulos is a postdoctoral research fellow at the Institution for Social and Policy Studies at Yale University. He is also visiting assistant professor in the department of political science at Fordham University, where he serves as director of the Elections and Campaign Management Program.

TABLE 1
Granger Causality Tests

 Rumsfeld Approval Bush Approval

Rumsfeld [approval.sub.t-1] 1.11 ** 0.23
 (0.29) (0.27)
Bush [approval.sub.t-1] -0.12 0.83 *
 (0.36) (0.35)
Constant 2.95 -2.11
 (7.02) (6.76)
Number of cases 13 13
Adjusted R-squared 0.93 0.92
Partial F 0.00 0.06
p-value for partial F 0.99 0.81

* p < 0.05. ** p < 0.01.

TABLE 2
Variables and Descriptive Statistics

Variable Mean SD Min. Max. Range

Rumsfeld approval 68.9 15.4 47 93 {0, 100}
Bush approval 64.9 13.0 51 90 {0, 100}
Unemployment (change) 0.04 0.25 -0.2 0.7 (-[infinity],
 +[infinity])
Inflation (change) 0.02 0.60 -0.8 1.1 (-[infinity],
 +[infinity])
Macropartisanship 48.1 2.68 45 55 {0, 100}
Non-scandal stories 748 295 484 1,403 {0,
 +[infinity])
Scandal stories 16 31 0 118 {0,
 +[infinity])
War casualties stories 752 330 290 1,309 {0,
 +[infinity])

Note: N = 14 for all variables. See Appendix for operationalizations.

TABLE 3
Determinants of Opinion toward Donald
Rumsfeld (Prais-Winsten Regression)

Dependent Variable: Rumsfeld Approval
Independent Variables Coefficient

Rumsfeld stories (non-scandal) 0.07 **
 (0.03)
Rumsfeld stories (scandal) -0.17 *
 (0.08)
U.S. casualties stories -0.05 **
 (0.02)
Unemployment (change) 13.65
 (17.17)
Inflation (change) 4.66
 (4.08)
Macropartisanship -1.49
 (0.89)
Constant 127.87 **
 (47.03)
[rho] -0.60
Number of cases 14
Adjusted R-squared 0.92
Durbin-Watson (transformed) 2.13

Note: Estimated errors are in parentheses.

* p < 0.10. ** p < 0.05.